Skip to content

Beyond Retrieval v2

Enterprise-grade Retrieval-Augmented Generation platform for document intelligence.

Upload documents. Ask questions. Get cited answers grounded in your actual content.

Get Started API Reference

What is Beyond Retrieval?

Beyond Retrieval v2 is a NotebookLM-inspired knowledge management system that transforms raw documents into searchable, AI-powered knowledge bases. Organize your knowledge into notebooks, ingest files in many formats, and have intelligent conversations grounded in your actual documents — with precise citations back to the source text.

Key Features

Document Ingestion

Upload PDF, DOCX, TXT, XLSX, CSV. Parse via Docling or Mistral OCR. Chunk with Recursive, Hybrid, or Agentic strategies. Embed with OpenAI, OpenRouter, or Ollama.

RAG Chat

Ask natural-language questions and get answers with numbered citations. 6 personas, 10 languages, smart language mode, cache-first pipeline with semantic similarity.

AI Enhancement

Kanban board for chunk-level context augmentation. Parallel processing with per-notebook concurrency control. Publish enhanced chunks to the vector store.

Search Playground

7 retrieval strategies: Fusion, Semantic, Full-Text, Cache, Contextual, Agentic, Smart Router. Side-by-side A/B comparison mode and search history.

LLM Judge

Background quality evaluation after every RAG response. Cache eligibility scoring. Response quality logging for continuous improvement.

Multi-Provider LLM

OpenRouter as gateway (25+ models), OpenAI Direct, Ollama (local/air-gapped). Per-notebook provider selection with dynamic API keys from the UI.

Flexible Deployment

Cloud Supabase or fully self-hosted local Supabase (Docker). 4 storage backends: Supabase Storage, S3, Local filesystem, None. 4 deployment modes.

Auth & Sharing

Clerk JWT authentication with bypass mode for development. Invite links for notebook sharing with role-based access control (admin / chat_only).

At a Glance

Metric Value
API Endpoints 83+ across 15 route modules
Unit Tests 1213+ with end-to-end coverage
LLM Providers 3 (OpenRouter, OpenAI, Ollama)
Retrieval Strategies 7
Supported Languages 10
Chat Personas 6
Deployment Modes 4
Docker Services 16 (across base + profiles)

Tech Stack

Layer Technology
Backend Python 3.12, FastAPI 0.115, Pydantic v2, Pydantic AI
Frontend React 19, Vite 7, Tailwind CSS 4, Lucide Icons
Database PostgreSQL 15 + pgvector (HNSW indexes), Supabase
Auth Clerk (JWT, JWKS verification) with bypass mode
LLM Providers OpenRouter, OpenAI, Ollama
Document Parsing Docling, Mistral OCR
Embeddings OpenAI text-embedding-3-small (1536d), 21+ OpenRouter models, Ollama nomic-embed-text
Storage Supabase Storage, S3, Local filesystem
Reverse Proxy Caddy 2 (auto-HTTPS via Let's Encrypt)
Containerization Docker Compose

Architecture

graph TB
    User[User / Browser]

    subgraph Caddy ["Caddy (Reverse Proxy, Auto-HTTPS)"]
        CaddyProxy[Caddy 2]
    end

    subgraph App ["Application"]
        Frontend["Frontend<br/>React 19 + Vite 7"]
        Backend["Backend<br/>FastAPI + Pydantic AI"]
    end

    subgraph Database ["Database Layer"]
        Supabase["Supabase<br/>PostgreSQL 15 + pgvector"]
        Storage["Supabase Storage<br/>/ S3 / Local FS"]
    end

    subgraph LLM ["LLM Providers"]
        OpenRouter["OpenRouter<br/>25+ models"]
        OpenAI["OpenAI Direct"]
        Ollama["Ollama<br/>Local / Air-Gapped"]
    end

    subgraph Parsing ["Document Parsing"]
        Docling["Docling Serve"]
        MistralOCR["Mistral OCR"]
    end

    User --> CaddyProxy
    CaddyProxy --> Frontend
    CaddyProxy --> Backend
    Backend --> Supabase
    Backend --> Storage
    Backend --> OpenRouter
    Backend --> OpenAI
    Backend --> Ollama
    Backend --> Docling
    Backend --> MistralOCR
    Frontend --> Backend

Next Steps